Honey Bee Image Dataset for Machine Learning and Computer Vision Model Building

Chaudhary, P., Foley, C., Samiappan, S., Kohler, L., Senyurek, V., Boes, D., & Chakrabarti, P. (2024). Honey Bee Image Dataset for Machine Learning and Computer Vision Model Building. GRI Publications and Scholarship. Mississippi State University: Geosystems Research Institute. DOI:10.54718/JCHW9401.

The dataset consists of 4,590 frames distributed across 15 different video recordings representing diverse lighting and seasonal conditions of honey bee activity. These frames were labeled in three classes: non pollen carrying worker honey bees, pollen carrying worker honey bees, and drone bees. The frames were extracted from videos captured using GoPro Hero 9 and Hero 11 cameras in a research apiary at Mississippi State University. The video recording process was accompanied with data collection of environmental factors such as temperature, humidity, wind direction and speed, solar radiation and other weather conditions. It provides contextual information related to bee behavior for every video recording. Annotation of the extracted frames from the dataset was done by a human expert, following strict guidelines to ensure precision and consistency. The dataset can be used by researchers for training honey bee detection computer vision models aimed at automating bee detection and classification for monitoring tasks as well as honey bee behavior analysis, providing insight into hive activity and foraging patterns.